Perilous Journeys to Better Lands : Exploring Undocumented Grave Yards

Visualization and Analysis

Problem Statement

People around the world are constantly trying to better their lives, with many far more fortunate than refugees, those forced to escape their homelands to avoid conflict, persecution or natural disaster. In a bid to flee these dire conditions, they inevitably travel using untrodden and dangerous routes, leading to many deaths. By looking at how migrant flows, as well as major points at which many deaths occur, are affected by the incident that precipitated the outward migration from the region and other contemporaneous events, it is hoped that future migrant flows can be predicted. By knowing where people will go, and where they are most likely to run into resistance or danger, aid agencies may be better equipped to deal with migrant crises around the world, giving these refugees a shot at a better life.

Data

The required data was obtained from two reliable websites- https://missingmigrants.iom.int/ and http://www.unhcr.org/ . From the first website, the most updated data of missing migrants in the world from January 2014 to March 2018 can be found. There is sufficient information for us to make detailed spatial visualizations, using data such as the number of dead and missing, filtered according to their geographical grouping, location, and cause of death. In addition, there is information on migrant routes, albeit not all migrant deaths contain this data.

The second source of data, the UNHCR database, contains accurate geospatial information on the movement of refugees. It contains the locations from which refugees leave and enter various territories, as well as the number and categories of people making the aforementioned movement. By cross-referencing to the first data source, the two data sources can be joined, providing a more complete image of migrant movements, thus obtaining more informative data that will be useful for visualization.

Analysis and Visualization

The primary purpose of this project is to present, as accurately as possible, the best available data on the movement of migrants and a global profile of hot zones and regional trends for migrant fatalities and disappearances. The project revolves around analysing data to understand human migration patterns, especially in the hot zones, to highlight the need to create safe passage for refugees, as opposed to the current perilous paths they take. This may be done by strengthening the capacity and preparedness of search and rescue operations in these hot zones.

Our visualization methodology will be an infographic poster (see link down) that includes descriptive statistics and spatial maps to showcase interesting trends.

Load Relevant Libraries

library(tidyverse)
library(ggplot2)
library(lubridate)
library(maps)
library(maptools)
library(ggraph)
library(igraph)
library(ggmap)
library(rworldmap)
library(sp)
library(readxl)
library(ggalluvial)

Data preprocessing

We will process the data obtained from https://missingmigrants.iom.int/ into a useable format. We quickly did an initial visualization using QGIS to get a sense of the data and noticed some incorrect data points which will be filtered out in this step.

unprocessed_events <- read_csv("MissingMigrants-Global-2018-04-07T23-25-35.csv")

processed_events_2 <- cbind(unprocessed_events, str_split_fixed(unprocessed_events$`Location Coordinates`, ",", 2))
names(processed_events_2)[21] <- "Latitude"
names(processed_events_2)[22] <- "Longitude"
processed_events_2$Longitude <- as.numeric(as.character(processed_events_2$Longitude))
processed_events_2$Latitude <- as.numeric(as.character(processed_events_2$Latitude))
events <- processed_events_2
events <- filter(events, events$Latitude > -60)

Visualizing the preprocessed data

map <- NULL
mapWorld <- borders("world", colour="gray50")
map <- ggplot() + mapWorld
basemap <- map
map <- map + geom_point(aes(x=events$Longitude, y=events$Latitude) ,color="red", size=1.5, alpha=0.2) 
map <- map + xlab("LONGITUDE") + ylab("LATITUDE") + ggtitle("Reported Migrant Death and Missing Events: 2014-2018")
map

The plot reveals clusters in reported events of missing or dead migrants around the Mediterranean, North African and US-Mexico regions. It is very strange to note that there are no migrant deaths/missing events reported around the South Eastern Indian Ocean closer to Australia and Papua New Guinea. We will address these issues in the conclusion.

Understanding initial statistics

We will quantify the data based on different criteria to get useful insights.

events %>%
  filter(!is.na(`Region of Incident`)) %>%
  group_by(`Region of Incident`) %>%
  summarise(count = n()) %>%
  ggplot(aes(x = `Region of Incident`, y = count)) + geom_bar(stat = "identity") + xlab("Region") + ylab("Number of Reported Events") +
  ggtitle("Number of Migrant Death Events reported by Region : 2014-2018")

events %>%
  filter(!is.na(`Reported Year`)) %>%
  group_by(`Reported Year`) %>%
  summarise(count = n()) %>%
  ggplot(aes(x = `Reported Year`, y = count)) + geom_bar(stat = "identity") + xlab("Year") + ylab("Number of Reported Events") +
  ggtitle("Number of Migrant Death Events reported by Year : 2014-2018")

events %>%
  filter(!is.na(`Region of Incident`)) %>%
  filter(!is.na(`Number Dead`)) %>% 
  group_by(`Region of Incident`) %>%
  summarise(total = sum(`Number Dead`)) %>%
  ggplot(aes(x = `Region of Incident`, y = total)) + geom_bar(stat = "identity") + xlab("Region") + ylab("Number of Fatalities") +
  ggtitle("Number of Migrant Death Toll reported by Region : 2014-2018")

events %>%
  filter(!is.na(`Reported Year`)) %>%
  filter(!is.na(`Number Dead`)) %>% 
  group_by(`Reported Year`) %>%
  summarise(total = sum(`Number Dead`)) %>%
  ggplot(aes(x = `Reported Year`, y = total)) + geom_bar(stat = "identity") + xlab("Year") + ylab("Number of Fatalities") +
  ggtitle("Number of Migrant Death Toll reported by Year : 2014-2018")

Based on the preliminary descriptive statistics, it is clear that the Mediterranean region and the associated North African region report the largest proportion of migrant death events and fatalities. Hence, we will localise our focus around the Mediterranean region from this point.

Mediterranean region at a Glance

We will quantify the migrant death and fatalities spatial data in the Mediterranean region to get some useful insights before proceeding to descriptive statistical study for this region.

map <- ggmap(get_googlemap(center = 'mediterranean',zoom=4, maptype = "satellite"))

med_events <- events %>%
  filter(`Region of Incident` == "Mediterranean")

med_events[1:10,]

##    Web ID Region of Incident  Reported Date Reported Year Reported Month
## 1   44693      Mediterranean April 01, 2018          2018            Apr
## 2   44694      Mediterranean April 01, 2018          2018            Apr
## 3   44692      Mediterranean March 29, 2018          2018            Mar
## 4   44676      Mediterranean March 20, 2018          2018            Mar
## 5   44671      Mediterranean March 18, 2018          2018            Mar
## 6   44669      Mediterranean March 17, 2018          2018            Mar
## 7   44665      Mediterranean March 15, 2018          2018            Mar
## 8   44661      Mediterranean March 12, 2018          2018            Mar
## 9   44672      Mediterranean March 08, 2018          2018            Mar
## 10  44648      Mediterranean March 03, 2018          2018            Mar
##    Number Dead Number Missing Total Dead and Missing Number of Survivors
## 1            5              6                     11                   1
## 2            1             NA                      1                  NA
## 3           NA              7                      7                  NA
## 4            1             NA                      1                  NA
## 5            1             NA                      1                  NA
## 6           16              3                     19                   3
## 7           12             NA                     12                  22
## 8            1             NA                      1                  NA
## 9            1             NA                      1                  NA
## 10          NA             21                     21                  30
##    Number of Females Number of Males Number of Children
## 1                 NA               1                 NA
## 2                 NA               1                 NA
## 3                 NA               7                 NA
## 4                 NA               1                 NA
## 5                 NA               1                 NA
## 6                  5              11                  9
## 7                 NA              NA                 NA
## 8                 NA               1                 NA
## 9                 NA               1                 NA
## 10                 4              17                 NA
##                              Cause of Death
## 1                                  Drowning
## 2                         Presumed drowning
## 3                         Presumed drowning
## 4                         Presumed drowning
## 5                                  Drowning
## 6                                  Drowning
## 7                         Presumed drowning
## 8  Sickness and lack of access to medicines
## 9                                  Drowning
## 10                        Presumed drowning
##                                                       Location Description
## 1  Unspecified location in the Gibraltar Strait, between Morocco and Spain
## 2                                     Off the coast of Al-Hoceima, Morocco
## 3  Unspecified location in the Gibraltar Strait, between Morocco and Spain
## 4                                Body found on the shore of Tripoli, Libya
## 5                          Body recovered on a beach in Rota, Cádiz, Spain
## 6                 Unspecified location off the coast of Agathonisi, Greece
## 7                                    Alboran Sea between Morocco and Spain
## 8   Hospital in Modica, Sicily, after being rescued off the coast of Libya
## 9                          Body recovered on a beach in Rota, Cádiz, Spain
## 10                                53 nautical miles off the coast of Libya
##                               Information Source
## 1       IOM Spain, Salvamento Marítimo, El Mundo
## 2                                  EFE, Yabiladi
## 3                            Caminando Fronteras
## 4                                      IOM Libya
## 5                    Guardia Civil, Europa Press
## 6  IOM Greece, Hellenic Coast Guard, Reuters, AP
## 7                            Caminando Fronteras
## 8                      Proactiva Open Arms, Ansa
## 9                    Guardia Civil, Europa Press
## 10                   SOS Méditerranée, IOM Italy
##                Location Coordinates       Migration Route
## 1  35.977554386570, -5.768318886719 Western Mediterranean
## 2  35.307342028997, -3.957839329297 Western Mediterranean
## 3  35.961994206451, -5.382424111328 Western Mediterranean
## 4  32.908397062094, 13.256741284619 Central Mediterranean
## 5  36.619003063120, -6.364435634424 Western Mediterranean
## 6  37.462855742504, 27.010840066992 Eastern Mediterranean
## 7  35.519970536120, -2.645016567187 Western Mediterranean
## 8  36.839922534108, 14.765837528836 Central Mediterranean
## 9  36.626580524694, -6.383661708643 Western Mediterranean
## 10 33.442744459548, 12.570127875000 Central Mediterranean
##                                                                                               URL
## 1  https://bit.ly/2EddpwF, https://bit.ly/2GS3tye, https://bit.ly/2uMxECe, https://bit.ly/2GzE74U
## 2                                                  https://bit.ly/2GtFQwv, https://bit.ly/2GTiMXz
## 3                                                                          https://bit.ly/2Ej7nL0
## 4                                                                                            <NA>
## 5                                                                           http://bit.ly/2FS0mql
## 6                     http://www.hcg.gr/node/17395, http://bit.ly/2u2YRjw, http://reut.rs/2u2ZdGS
## 7                                                                           http://bit.ly/2FDRbtf
## 8                                                    http://bit.ly/2tIL5m3, http://bit.ly/2DmKA0w
## 9                                                                           http://bit.ly/2DMnWPj
## 10                            http://bit.ly/2H4VLgy, http://bit.ly/2oMrI6u, http://bit.ly/2oQ5GQx
##    UNSD Geographical Grouping Verification level Latitude Longitude
## 1               Uncategorized                  5 35.97755 -5.768319
## 2               Uncategorized                  3 35.30734 -3.957839
## 3               Uncategorized                  4 35.96199 -5.382424
## 4               Uncategorized                  4 32.90840 13.256741
## 5               Uncategorized                  5 36.61900 -6.364436
## 6               Uncategorized                  5 37.46286 27.010840
## 7               Uncategorized                  4 35.51997 -2.645017
## 8               Uncategorized                  4 36.83992 14.765838
## 9               Uncategorized                  5 36.62658 -6.383662
## 10              Uncategorized                  4 33.44274 12.570128

map + geom_point(aes(x=Longitude, y=Latitude), data=med_events, col="red", size = 2) + ggtitle("Mediterranean Migrant Deaths Events: 2014 - 2018")

map  + ggtitle("Mediterranean Migrant Deaths Events: 2014 - 2018") + geom_density2d(aes(x=Longitude, y=Latitude, colour = 'red'), data=med_events)

map + geom_point(alpha = 0.05, aes(x=Longitude, y=Latitude), data=med_events, col="red", size = med_events$`Number Dead`/10) + scale_size_area() + ggtitle("Mediterranean Migrant Fatalities: A proportional symbol map\n2014 - 2018")

# This is similar to the plot above, but with Total Dead and Missing instead of Number Dead.
map + geom_point(alpha = 0.4, aes(x=Longitude, y=Latitude, size=med_events$`Total Dead and Missing`, colour=med_events$`Migration Route`), data=med_events) + scale_size_area() + scale_fill_brewer(palette = 3) + guides(color=guide_legend(title="Region"), size=guide_legend(title = "Total Dead and Missing")) +  ggtitle("Mediterranean Migrant Missing and Fatalities Toll: A proportional symbol map\n2014 - 2018")

Mediterranean - Understanding statistics

We will classify and quantify the data based on different regions of the Mediterranean and different times of the year.

# We used `factor(match(`Reported Month`, month.abb))` to order the months.
med_events %>% 
  mutate(month_number = match(`Reported Month`, month.abb)) %>%
  ggplot(aes(x=factor(month_number))) + geom_bar() + facet_wrap(~`Migration Route`) + xlab("Reported Month")

med_events %>% 
  ggplot(aes(x=`Reported Year`)) + geom_bar() + facet_wrap(~`Migration Route`)

med_events %>%
  mutate(Date=mdy(med_events$`Reported Date`)) %>%
  ggplot(aes(x=Date)) + geom_histogram(binwidth=28) 

You might have wondered why we are not focusing on time-series analysis (by year). The primary reason is that with time the technology, awareness and methods used to report migrant death and missing events has increased. i.e.: Though we have more migrant death and missing events reported in 2015 compared to 2014, it does not mean that 2014 actually had less migrant death and missing events, but it is primarily because we have better methods and technology to track these events in 2015 compared to 2014. So time-series analysis will not yield correct insights.

The useful information that can be obtained here is that the Central Mediterranean is the primary actor in migrant deaths and missing toll. We will include spatial maps below to see this better.

map + geom_point(alpha = 0.4, aes(x=Longitude, y=Latitude, size=med_events$`Total Dead and Missing`, colour="red"), data=med_events) + scale_size_area(name="Total dead and missing") + facet_wrap(~`Reported Month`) + scale_colour_discrete(name=NULL, breaks=NULL) 

map + geom_point(alpha = 0.4, aes(x=Longitude, y=Latitude, size=med_events$`Total Dead and Missing`, colour="red"), data=med_events) + scale_size_area(name="Total dead and missing") + facet_wrap(~`Reported Year`) + scale_colour_discrete(name=NULL, breaks=NULL)

An assumption

The data of missing migrants that we are using does not contain any information on the nationality of the victims. So we are not able to exclusively formulate any methodology to understand trends between country of origin and missing migrants. So we are making an assumption that, Proportion of fatalities where victims belong to country Z = proportion of country Z asylum seekers in the EU. To give a concrete example, if 20% of asylum seekers in the EU are from Iraq, then we assume that 20% of people who die in the Mediterranean are Iraqis.

Based on the above assumption we are trying to visualize any trends in migrant flow and fatalities.

Preprocess asylum seekers data

asylum_seekers <- read_csv("migr_asylum_Data.csv")

# Disable scientific notation
options(scipen=999)

# Get total number of asylum seekers and event death and missing count
total_asylum_seekers <- sum(asylum_seekers$Value, na.rm = TRUE)
total_asylum_seekers

## [1] 7159355

total_dead_and_missing <- sum(med_events$`Total Dead and Missing`, na.rm = TRUE)
total_dead_and_missing

## [1] 15865

# Get asylum seeker count by country
asylum_seekers.country <- asylum_seekers %>%
  select(CITIZEN, GEO, Value) %>%
  group_by(asylum_seekers$CITIZEN) %>%
  filter(!is.na(Value)) %>%
  summarise(asylum_seekers_count = sum(Value))

asylum_seekers.country <- asylum_seekers.country %>%
  mutate(dead_and_missing_count = asylum_seekers_count*total_dead_and_missing/total_asylum_seekers)

asylum_seekers.country[1:10,]

## # A tibble: 10 x 3
##    `asylum_seekers$CITIZEN` asylum_seekers_count dead_and_missing_count
##    <chr>                                   <dbl>                  <dbl>
##  1 Algeria                                119800                265.   
##  2 Angola                                  14870                 33.0  
##  3 Bahrain                                   535                  1.19 
##  4 Benin                                    7200                 16.0  
##  5 Botswana                                  285                  0.632
##  6 Burkina Faso                            13775                 30.5  
##  7 Burundi                                  6930                 15.4  
##  8 "C\xf4te d'Ivoire"                     109580                243.   
##  9 Cameroon                                53870                119.   
## 10 Cape Verde                                 70                  0.155

Sankey diagrams as a means to understand Migrant flow

Sankey diagrams can create visual emphasis on major flows within our system, therefore we have chosen it to represent trends in migrant flows.

Non-spatial representation of migrant flow

After preprocessing the asylum seekers data, we have identified the most popular origin and host countries and we will include some interesting statistics below.

asylum_seekers %>% 
  filter(Value > 0) %>% 
  filter(GEO %in% c("Germany (until 1990 former territory of the FRG)",
                    "Italy",
                    "Sweden",
                    "France",
                    "Hungary",
                    "Austria",
                    "Greece",
                    "Netherlands",
                    "Switzerland",
                    "United Kingdom",
                    "Belgium")) %>% 
  filter(CITIZEN %in% c("Syria",
                        "Iraq",
                        "Eritrea",
                        "Nigeria",
                        "Somalia",
                        "Gambia, The",
                        "Guinea",
                        "Mali",
                        "Sudan",
                        "Algeria",
                        "Senegal",
                        "Cote d'Ivoire",
                        "Morocco",
                        "Egypt",
                        "Libya"
                        )) %>% 
  select(CITIZEN, GEO, Value) %>% 
  mutate(GEO = replace(GEO, GEO == "Germany (until 1990 former territory of the FRG)", "Germany")) %>%
  group_by(CITIZEN, GEO) %>% 
  summarise(NumberOfAsylumSeekers = sum(Value)) %>% 
  arrange(desc(NumberOfAsylumSeekers))

## # A tibble: 154 x 3
## # Groups:   CITIZEN [14]
##    CITIZEN     GEO         NumberOfAsylumSeekers
##    <chr>       <chr>                       <dbl>
##  1 Syria       Germany                    527735
##  2 Iraq        Germany                    164245
##  3 Syria       Sweden                      93270
##  4 Nigeria     Italy                       82660
##  5 Syria       Hungary                     77030
##  6 Eritrea     Germany                     54725
##  7 Syria       Austria                     49200
##  8 Syria       Greece                      48580
##  9 Gambia, The Italy                       35090
## 10 Syria       Netherlands                 33805
## # ... with 144 more rows

asylum_seekers %>% 
  filter(Value > 0) %>% 
  filter(GEO %in% c("Germany (until 1990 former territory of the FRG)",
                    "Italy",
                    "Sweden",
                    "France",
                    "Hungary"
                    )) %>% 
  filter(CITIZEN %in% c("Syria",
                        "Iraq",
                        "Eritrea",
                        "Nigeria",
                        "Somalia",
                        "Gambia, The",
                        "Guinea",
                        "Mali",
                        "Sudan",
                        "Algeria",
                        "Senegal",
                        "Cote d'Ivoire"
                        )) %>% 
  select(GEO, Value) %>%
  mutate(GEO = replace(GEO, GEO == "Germany (until 1990 former territory of the FRG)", "Germany")) %>% 
  group_by(GEO) %>% 
  summarise(sumValue = sum(Value)) %>% 
  arrange(desc(sumValue)) %>% 
  ggplot(aes(x = reorder(GEO, -sumValue), y = sumValue, fill=factor(GEO) )) +# factor makes it non continuous (categorical)
  geom_bar(stat="identity") + scale_fill_brewer(palette="Dark2") +
  xlab("Destination") + ylab("Number of asylum seekers") +
  ggtitle("Top 5 Asylum Seeking Destinations in the EU") +
  labs(fill = "Destination") +
  theme (axis.text.x = element_text(angle = 45, hjust = 1), plot.title = element_text(hjust = 0.5), legend.position = "none")

Asylum Seekers Movement: A visualization in the non-spatial domain

asylum_seekers %>% 
  filter(Value > 200) %>% 
  filter(GEO %in% c("Germany (until 1990 former territory of the FRG)",
                    "Italy",
                    "Sweden",
                    "France",
                    "Hungary",
                    "Austria",
                    "Greece",
                    "Netherlands",
                    "Switzerland",
                    "United Kingdom",
                    "Belgium")) %>% 
  filter(CITIZEN %in% c("Syria",
                        "Iraq",
                        "Eritrea",
                        "Nigeria",
                        "Somalia",
                        "Gambia, The",
                        "Guinea",
                        "Mali",
                        "Sudan",
                        "Algeria",
                        "Senegal",
                        "Cote d'Ivoire")) %>% 
  select(CITIZEN, GEO, Value) %>% 
  mutate(GEO = replace(GEO, GEO == "Germany (until 1990 former territory of the FRG)", "Germany")) %>% 
  ggplot(aes(weight = Value, axis1 = CITIZEN, axis2 = GEO)) +
  geom_alluvium(aes(fill = CITIZEN), width = 1/6) +
  geom_stratum(alpha = .5,  width = 1/6, color = "black") +
  geom_text(stat = "stratum", label.strata = TRUE, size = 3) +
  scale_x_continuous(breaks = 1:2, labels = c("Origin", "Destination")) +
  ggtitle("Asylum Seekers Movement") +
  theme(axis.title.y=element_blank(),
        axis.text.y=element_blank(),
        axis.ticks.y=element_blank(),
        legend.position = "none")

Spatial representation of migrant flow

We will understand the migrant flow by incorporating the underlying geographical layer in this section. First we extract our filtered data and we create a graph from it. Do note that Edge.csv and Nodes.csv were manually created by us. In this part we will walk through on how we created our Sankey diagrams and we will show a concrete example as well(Germany).

edges <- read.csv("Edge.csv", sep=";")
nodes <- read.csv("Nodes.csv", sep=";")
edges <- edges[,1:3]
countryGraph <- graph_from_data_frame(edges)

# Allows only nodes that are in edges of the graph.
nodes <-
  filter(nodes, nodes$Country %in% edges$From | nodes$Country %in% edges$To)

# Orders nodes to be the same as graph's Vertex's order
arrangedNodes <- nodes[order(match(nodes$Country, V(countryGraph)$name)),]

# After ordering, we are able to assign Longitude and Latitude by index instead of name
V(countryGraph)$x = arrangedNodes$Longitude
V(countryGraph)$y = arrangedNodes$Latitude

# We can select a particular country this way. 
E(countryGraph)[inc(V(countryGraph)[name=="Germany"])]

## + 12/151 edges from bbbeaee (vertex names):
##  [1] Syria        ->Germany Iraq         ->Germany Eritrea      ->Germany
##  [4] Nigeria      ->Germany Somalia      ->Germany Gambia, The  ->Germany
##  [7] Guinea       ->Germany Mali         ->Germany Sudan        ->Germany
## [10] Algeria      ->Germany Senegal      ->Germany Cote d'Ivoire->Germany

# This is a subgraph for an individual country
subGraph <- subgraph.edges(countryGraph, E(countryGraph)[inc(V(countryGraph)[name=="Germany"])]) 

# We then sqrt all the values to decrease them to a more aesthetically visible value while still maintaining their relative sizes
E(subGraph)$Value = sqrt(E(subGraph)$Value)

# Plots graph
# error "data must be uniquely named but has duplicate elements" would appear if we are using the devtools version of ggplot2.
# We used the CRAN version instead.
subGraph %>%
  ggraph(layout='nicely') +
  geom_node_text(aes(label=name)) +
  geom_edge_diagonal(aes(width=Value/60, alpha=0.4), 
                     end_cap=rectangle(20,8, "mm"), 
                     start_cap=rectangle(20,3, "mm")) + 
  xlim(c(-20,50)) + ylim(c(0,70)) + scale_edge_width_identity() + theme_void() + theme(legend.position="none")

We make use of ggmap to add the underlying spatial data.

worldMap <- get_googlemap(center=c(15,35),zoom=3,style="element:labels|visibility:off", maptype="hybrid")
ggmap(worldMap, base_layer=ggraph(subGraph, layout='auto')) +  
  geom_node_text(aes(x=x, y=y, label=name), size=3, colour="red") +
  geom_edge_diagonal(aes(width=Value/60, alpha=0.4), colour="white") + 
  xlim(c(-20,50)) + ylim(c(0,70)) + scale_edge_width_identity() + theme_void() + theme(legend.position="none")

Now to obtain multiple plots, we can use a for loop to change the filtered country every iteration. We also make use of ggsave to save our plots. You can find all the plots at https://github.com/keshik6/Perilous_Journey_to_Better_Lands. The sankey diagrams are shown in the poster as well.

# This loop will create the sankey diagrams without the underlying spatial layer. The code for this is included as a reference only. The proper final plots will be created in the following part and saved.
for (i in arrangedNodes$Country) {
  subGraph <- subgraph.edges(countryGraph, E(countryGraph)[inc(V(countryGraph)[name==i])]) 
  E(subGraph)$Value = sqrt(E(subGraph)$Value)
  ggsave(str_c("Maps\\",i, ".pdf"), device="pdf",width=10,height=10,plot=
    subGraph %>%
      ggraph(layout='auto') +
      geom_node_text(aes(x=x, y=y, label=name)) +
      geom_edge_diagonal(aes(width=Value/60, alpha=0.4), 
                         end_cap=rectangle(20,8, "mm"), 
                         start_cap=rectangle(20,3, "mm")) + 
      xlim(c(-20,50)) + ylim(c(0,70)) + scale_edge_width_identity() + theme_void() + theme(legend.position="none")
  )
}

## for loop with country plots
p <- vector( length=nrow(arrangedNodes))
for (i in arrangedNodes$Country) {
  subGraph <- subgraph.edges(countryGraph, E(countryGraph)[inc(V(countryGraph)[name==i])]) 
  E(subGraph)$Value = sqrt(E(subGraph)$Value)
  j <-ggmap(worldMap, base_layer=ggraph(subGraph, layout='auto')) +  
     geom_node_text(aes(x=x, y=y, label=name), size=3, colour="white") +
     geom_edge_diagonal(aes(width=Value/60, alpha=0.4, colour="red")) + 
     xlim(c(-20,50)) + ylim(c(0,70)) + scale_edge_width_identity() + theme_void() + theme(legend.position="none") 
  ggsave(str_c("Maps_hybrid\\",i, ".pdf"), device="pdf",width=10,height=10,plot=j)
}

If you observe these Sankey diagrams for migrant flow through the Mediterranean Sea it is quite interesting to note that the thickness/visual emphasis on the flow of the arrows that end up in the Mediterranean Sea is very minute. That is the number of fatalities per country normalized based on the total number of people of the particular country who successfully seek asylum in the EU is insignificant. Our analysis becomes very interesting at this point. Though the above proportions appear very small numerically we have more than 22500 migrant death and missing toll from 2014 - 2018 in the Mediterranean region. Do the authorities and governments understand the seriousness of this situation or does the percentages behind being small allow them to consider rescue operations as trivial relative to the investment of money and labour that these operations incur?

Mediterranean Search and Rescue(SAR) Operations

SAR = data.frame(read_excel("SAR Naval Operation.xlsx"))
SAR %>% 
  group_by(Year) %>% 
  summarise(`Total Rescued People` = sum(Rescued.People)) %>% 
  ggplot(aes(x = Year, y = `Total Rescued People` )) +
  geom_line()+
  geom_point() +
  ggtitle("Number of Rescued People in Meditteranean Sea") +
  theme (plot.title = element_text(hjust = 0.5)) +
  labs(fill = "Year")

It is quite useful to visualize the number of rescued people by different Naval Units in the Mediterranean Sea from 2014-2018.

ggplot(data = SAR, aes(x = Naval.Unit, y = (Rescued.People), fill = factor(Year))) +   
  geom_bar(position="dodge", stat="identity") + xlab("Naval Unit") + ylab("Number of Rescued People") +
  ggtitle("Number of Rescued People by Naval Units in Meditteranean Sea") +
  theme (axis.text.x = element_text(angle = 45, hjust = 1), plot.title = element_text(hjust = 0.5)) +
  labs(fill = "Year")

There is a large drop in the number of rescued people in the Mediterranean from 2016 to 2017. We do not have sufficient evidence to explain the reasoning behind this.

Conclusion

In the requirement formulation phase of the project, one of our key deliverable was developing predictive models of hot zones by analyzing the shifts in hot zones with time from the available data. But due to insufficient data and limited expertise on the topic we were not able to complete this requirement. So our requirements were reframed to make the project rich in visualization and limit the scope of our analysis.

We would like to reiterate some interesting insights, findings and concerns at this point. First and foremost, it was very strange and surprising to note that no data on migrant death and missing cases were reported in the South Eastern Indian Ocean region (around Australia and Papua New Guinea). This is something that needs to be addressed by the respective authorities and governments. Next interesting observation based on our assumption is the number of fatalities per country normalized based on the total number of people of the particular country who successfully seek asylum in the EU is insignificant. This can potentially lead to the seriousness of this issue being misunderstood by the respective authorities and governments. This could be the reason for the number of people rescued in the Mediterranean (by Naval Units) dropping by about 60% from 2016 to 2017. The rest of the observations do not result in any concrete conclusions but are able to explain us the seriousness of the migrant crisis in the Mediterranean region.

Concluding, we appreciate the opportunity and we look forward to extend our project and create interactive Sankey diagrams to visualize migrant flow in the near future.